That loads all the data into memory, though. Fine if you're joining all of it, but with the Postgres example you can throw the whole SQL language at the data, and it will be streamed and pipelined efficiently just like with normal Postgres tables.